A comparison of authorship attribution approaches applied on the Lithuanian language

Abstract

This paper reports comparative authorship attribution results obtained on the Internet comments of the morphologically complex Lithuanian language. We have explored the impact of machine learning and similarity-based approaches on the different author set sizes (containing 10, 100, and 1,000 candidate authors), feature types (lexical, morphological, and character), and feature selection techniques (feature ranking, random selection). The authorship attribution task was complicated due to the used Lithuanian language characteristics, nonnormative texts, an extreme shortness of these texts, and a large number of candidate authors. The best results were achieved with the machine learning approaches. On the larger author sets the entire feature set composed of word-level character tetra-grams demonstrated the best performance.